Model observer design for detecting multiple abnormalities in anatomical background images

摘要: As psychophysical studies are resource-intensive to conduct, model observers commonly used assess and optimize medical imaging quality. Existing were typically designed detect at most one signal. However, in clinical practice, there may be multiple abnormalities a single image set (e.g., multifocal multicentric breast cancers (MMBC)), which can impact treatment planning. Prevalence of signals different across anatomical regions, human do not know the number or location priori. new techniques have potential improve multiple-signal detection digital tomosynthesis more effective for diagnosis MMBC than planar mammography), quality assessment approaches addressing such tasks needed. In this study, we present model-observer mechanism same dataset. To handle high dimensionality images, novel implementation partial least squares (PLS) was developed estimate sets efficient channels directly from images. Without any prior knowledge background signals, PLS capture interactions between provide discriminant information. Corresponding linear decision templates employed generate both image-level location-specific scores on presence signals. Our preliminary results show that observer using channels, compared our first attempts with Laguerre-Gauss achieve performance reasonably small optimal design vary as interest change.